Architecture for Early Detection of Age-Related Macular Degeneration Using Data Augmentation and Vision Transformers (ViT)

Abstract

Age-Related Macular Degeneration (AMD) is a leading cause of blindness among the geriatric population worldwide. While there is no definitive cure for this pathology, early identification of AMD allows for effective treatment administration. This paper introduces an architecture for a specialized module to identifying AMD at various evolutionary stages. An in-depth analysis was conducted examining investigations related to detecting age-related macular degeneration (AMD), with a focus on studies utilizing deep learning techniques and vision transformers. It is important to emphasize that most of these works have only addressed binary disease detection. Our initiative, incorporates an architecture that emphasizes data augmentation in the training set and utilizes the ViT vision transformer for analyzing retina images. The main aim is to attain a differentiated categorization (non-AMD, mild, moderate, and advanced) that serves as a basis for diagnosing AMD. The ViT trained model has shown 96.55% accuracy in this classification. Hence, it can be inferred that the outlined module holds noteworthy value as an additional support tool for ophthalmologists in the precise detection of Age-Related Macular Degeneration.